Get started
Known issues
What’s known not to work, or not checked yet, in the current version, and what to do instead. Something else wrong?
Run ./gunkslab doctor, try ./gunkslab down && ./gunkslab up, then see
Troubleshooting. If it’s still wrong, open an issue at
gitlab.com/gunkslab/desk-issues.
0.1.0 (2026-11-14)
- Windows 11 and Debian haven’t been run yet. macOS (Apple’s
container, and Docker Desktop), Ubuntu 24.04 and Rocky Linux 9 have had all ten steps of the install checklist, on machines that already had the runtime. Driving on the page was checked by a script in a headless browser, not by a person. 0.1.0 ships without a Windows run: Install’s Windows steps follow Docker’s and Microsoft’s pages, not tried. Install’s Debian steps (podman) are checked against Debian 12’s packages, not run end to end. - The page doesn’t open by itself on some machines.
./gunkslab upopens the browser withopen,xdg-openorwslview, whichever is there. With none of them (a Linux server, or WSL withoutwslu), open http://127.0.0.1:8080/ yourself../gunkslab doctorsays which it is, and on a Linux machine with no screen (over SSH, say) it says so.upprints the address either way. - A port held by another program isn’t always named.
./gunkslab doctornames the program holding 1883, 9001 or 8080 usinglsof, orsson Linux.ssonly names your own programs; for anyone else’s it says “another program”, andsudo ss -ltnpshows it. - A go-to can end
blockedbeside a gap it could have used. The station remembers what a bot’s distance sensor finds, goes round it, and looks round before giving up (protocol). But it stores what was found as discs 4 cm across with room around them, and keeps gl_zero clear as the circle round its corners, 18.3 cm across. At gl_zero’s size, that closes gaps beside boxes the bot would fit through. It’s how the station models things, not the sensor: in the simulator, given a perfect view stored the same way, the same go-tos endblocked. Drive the bot through by hand. - gl_zero can drive into something low at the side of its path. Its sensor’s view is 27° wide, as wide as the bot
only about 22 cm ahead, and level out there, 33 mm up. Something low at the very edge of the bot’s path isn’t seen,
nearer or further. For a 2 cm box that’s about the outer 4 mm at the tyres; lower things, more. In the simulator,
a 2 cm box placed like that is hit (
low-box-beside-the-path). This is the sensor’s limit, not something the station can work round. Keep low things a centimetre or two clear of a bot’s path, or drive it there by hand. - The firmware’s sensor regions haven’t been checked on a real sensor. The firmware reads gl_zero’s sensor as five regions, as the simulator does. Which of the chip’s cells look down, up, left and right, and whether a reading comes from the region it was aimed at, are worked out from ST’s manual, not tried. The first wired gl_zero gets the bench check in A real bot on the bench. Until then, its reflex may stop it at the wrong things.